Smartphone application processor chips incorporate numerous processor cores, typically including multiple CPU cores, GPUs, DSPs, video processors, and image signal processors. Considering all the processing power available in these chips, why does Motorola's recently introduced Moto X smartphone use a stand-alone DSP processor chip? And why would Motorola use a DSP based on a 10-year-old architecture?
The answer is that the Moto X uses this low-power DSP chip to provide "always-on" voice activation--a feature that requires modest performance, but extreme energy efficiency. As Brian Klug of AnandTech writes in his excellent review:
Moto X includes another custom feature called touchless control, which enables always-on voice recognition activation on the Moto X…Say "OK Google Now" with the phone in any state, even screen off (standby with screen off, powered on with screen on, just not fully turned off) and it'll activate and give you a voice prompt.
The Texas Instruments C55x DSP chip used in the Moto X was not chosen for its speed (which is rather modest), but for its energy efficiency -- that is, the ability to deliver sufficient performance with minimum power consumption. This explains why a 10-year-old processor architecture (manufactured in a modern fabrication process) can be a winning candidate for a new design.
Voice activation seems to be going over well with consumers. According to market research firm Argus Insights, consumers are embracing the Moto X, and responding positively to the voice activation feature.
Considering the economics of integration in digital ICs, future smartphones that use a dedicated DSP for voice activation will probably integrate the DSP into the application processor SoC, rather than using a separate chip. Either way, it's interesting to see DSPs with relatively low performance finding new uses alongside higher-performance processing engines. And whether it's a separate chip or a core integrated into the main SoC, using an additional processor to save energy is an example of the trend of using more transistors to save energy, which I wrote about in 2011.
As smartphones and other devices gain more sensors, there will be many opportunities to add useful functionality through "always on" capabilities, extending well beyond voice activation. For example, the Moto X also provides a feature called "active display." Quoting again from Brian Klug's review:
Pull the phone out of your pant or suit pocket for example, and the active display interface lights up. Flip it from face down to face up, and it will light up. If you have a bag or purse, the same applies. Leave it face up on a table, and it will breathe with the status indicator periodically.
In this case, the phone uses a dedicated low-power microcontroller chip to process signals from multiple sensors (accelerometer, gyro, and ambient light sensor) and determine when to activate the display.
Sometimes, to maximize responsiveness while minimizing power consumption, it will make sense to use sensors in a hierarchical fashion, with a simpler, low-power sensor keeping watch and activating a more expressive, higher-power sensor when appropriate. For example, a very low power ultrasonic sensor could detect a hand gesture near the device, and then wake up a camera (and associated processor) for detection of longer-range and more complex gestures.
Coupled with appropriate algorithms, the proliferation of sensors -- whether they're image sensors, accelerometers, ultrasonic, or other types -- is enabling designers to create devices that are more intelligent and responsive. Executing those algorithms with maximum energy efficiency will require thoughtful use of heterogeneous processor architectures, as illustrated by the Moto X.